README

Code for "LLM-OREF: An Open Relation Extraction Framework Based on Large Language Models"

1. Install Environment:

conda env create -f env.yml 

2. Data Format:

{
    "text": ...,
    "head": ...,
    "tail": ...,
    "label": ...,
    "ins_idx": ...
}

3. Train RP:

torchrun --nproc_per_node=${n} --master_port=${port}  src/train_RP.py \
  --model_name ${model_name} \
  --model_path ${model_path}\
  --train_path ${train_path} \
  --output_path ${output_path} \
  --few_shot ${few_shot} \
  --train_batch_size ${train_batch_size} \
  --gradient_accumulation_steps ${gradient_accumulation_steps} \
  --num_train_epochs ${num_train_epochs} \
  --max_grad_norm ${max_grad_norm} \
  --bf16 \
  --deepspeed ${config} \
  --logging_steps ${logging_steps} \
  --lr ${lr} \
  --lora_r ${lora_r} \
  --lora_alpha ${lora_alpha} \
  --lora_dropout ${lora_dropout} \
    

4. Train RD:

torchrun --nproc_per_node=${n} --master_port=${port}  src/train_RD.py \
  --model_name ${model_name} \
  --model_path ${model_path} \
  --train_path ${train_path} \
  --output_path ${output_path} \
  --RP_lora_path ${RP_lora_path} \
  --few_shot ${few_shot} \
  --temperature ${temperature} \
  --alpha ${alpha} \
  --train_batch_size ${train_batch_size} \
  --gradient_accumulation_steps ${gradient_accumulation_steps} \
  --num_train_epochs ${num_train_epochs} \
  --max_grad_norm ${max_grad_norm} \
  --bf16 \
  --deepspeed ${config} \
  --logging_steps ${logging_steps} \
  --lr ${LR} \
  --lora_r ${lora_r} \
  --lora_alpha ${lora_alpha} \
  --lora_dropout ${lora_dropout} \

5. Relation Discovery:

python src/relation_discovery.py \
  --model_path ${model_path} \
  --train_path ${train_path} \
  --test_path ${test_path} \
  --unknown_r_path ${unknown_r_path} \
  --discoverer_lora_path ${discoverer_lora_path} \
  --output_path ${output_path} \
  --K ${K} \
  --few_shot ${few_shot} \

6. Relation Denoising:

python src/relation_denoising.py \
  --model_path ${model_path} \
  --test_path ${test_path} \
  --RP_lora_path ${RP_lora_path} \
  --unknown_r_path ${unknown_r_path} \
  --output_path ${output_path} \
  --few_shot ${few_shot} \

7. Relation Prediction:

python src/relation_prediction.py \
    --model_path ${model_path} \
    --test_path ${test_path} \
    --demo_path ${demo_path} \
    --unknown_r_path ${unknown_r_path} \
    --RP_lora_path ${RP_lora_path} \
    --output_path ${output_path} \
    --few_shot ${few_shot} \